Overview
Zanwen (Ryan) Fu is a Software Engineer and M.S. Computer Science (AI/ML) student at Duke University, focused on building production-grade agentic AI systems — from multi-agent orchestration and LLM tooling to the distributed backends that enable reliable operation at scale.
He is the sole founder and engineer of VYNN AI, an agentic financial analyst platform built end-to-end as a single engineer: LangGraph multi-agent orchestration, real-time market data pipelines, and autonomous LLM-driven equity research with strict semantic-symbolic separation and deterministic financial validation — 50K+ lines of production code.
Previously, he designed core components of AutoCodeRover, an autonomous code repair system acquired by Sonar, building the JetBrains IDE plugin end-to-end and enhancing the agentic repair backend with a Self-Fix Agent and interactive feedback loop — lifting SWE-bench Verified to 51.6%. Sonar's Foundation Agent, built on the AutoCodeRover core, has since reached 79.2%, ranking #1 on the SWE-bench leaderboard (Feb 2026). He is also the sole first author of LUMINA, a four-agent LLM framework for automated citation screening in medical systematic reviews, achieving 98.2% sensitivity across 15 SRMAs (~150K citations) with 35x fewer missed studies vs. prior baselines.
His current research interest is agent harness design — the infrastructure layer that makes autonomous systems observable, recoverable, and trustworthy enough to run unsupervised. He writes about this at zanwenfu.com/blog.